To make large language models more reliable, researcher Athul Paul Jacob at MIT helped devise the consensus game, where an LLM’s subsystems are driven toward finding an answer they can agree on. Here’s how to play: quantamagazine.org/game-theory-ca…